AN INVESTIGATION OF THE ASSOCIATIONS AMONG RECOVERY, KEY ILLNESS CHARACTERISTICS AND BONE MINERAL DENSITY IN WOMEN WITH A HISTORY OF ANOREXIA NERVOSA

Transcription

1 AN INVESTIGATION OF THE ASSOCIATIONS AMONG RECOVERY, KEY ILLNESS CHARACTERISTICS AND BONE MINERAL DENSITY IN WOMEN WITH A HISTORY OF ANOREXIA NERVOSA by Esther J. Waugh A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Health Policy, Management and Evaluation University of Toronto Copyright by Esther J. Waugh (2009)

2 AN INVESTIGATION OF THE ASSOCIATIONS AMONG RECOVERY, KEY ILLNESS CHARACTERISTICS AND BONE MINERAL DENSITY IN WOMEN WITH A HISTORY OF ANOREXIA NERVOSA Esther J Waugh Doctor of Philosophy 2009 Graduate Department of Health Policy, Management and Evaluation University of Toronto Background: Reduced bone mineral density (BMD) is an established complication of anorexia nervosa (AN). There is inconclusive evidence as to whether this reduction in bone mass is permanent or can be reversed with recovery from AN. The objectives of this study were to: i. determine the extent of reversal of skeletal deficits with recovery from AN, and the duration of recovery required for complete reversal, if this occurred; and, ii. evaluate the effect of key illness characteristics on BMD. Methods: Women (aged years) who had previously received inpatient treatment for AN at one of two hospital-based programs were selected for this cross-sectional study; 514 healthy premenopausal women recruited from the community served as a control group. A detailed lifetime illness history was obtained by a Life History Calendar interview. BMD was measured by dual-energy X-ray absorptiometry (DXA) at the spine, hip and total body. Low BMD was defined as a weight and age-matched standard deviation (Z-score) of -1.5 at one or more skeletal sites. Participants were considered recovered if they had maintained a body mass index 18.5 kg/m 2 and resumed regular menstruation for 1 year. ii

3 Results: Of 190 AN participants, 77 were considered recovered and 113 were ill. The prevalence of low BMD was 11.7% in the recovered group, 47.3% in the ill group and 6.8% in the control group. The odds of low BMD in the recovered participants was significantly lower than in the ill participants (odds ratio [OR] = 0.17, 95% CI 0.07, 0.36, p<0.0001) and was not significantly different from the controls (OR = 1.81, 95% CI 0.79, 3.78, p=0.15). Duration of illness was associated with low BMD (OR = 1.16, 95% CI 1.08, 1.25, p<0.0001) and was negatively associated with the odds of AN recovery. Normal mean BMD values at each skeletal site were observed in women recovered 3 years. Conclusion: The results emphasize the importance of early and sustained AN recovery for the prevention and treatment of low bone mass in this population and may offer motivation for AN patients to make positive behavioural changes leading to successful, long-term recovery. iii

4 ACKNOWLEDGEMENTS First and foremost, I would like to thank my supervisor, Dr. Gillian Hawker, for all your time, effort and expertise that you have given to this project, and for being so patient with me over a rather lengthy PhD experience! I also thank the other members of my thesis committee: Dr. Blake Woodside, Dr. Dorcas Beaton and Dr. Pierre Coté, for your guidance and thoughtful reviews of my work. You were a wonderful team to work with. I would also like to thank Ruth Croxford, who I have actually never met, but who always provided prompt and much-needed statistical advice whenever asked. Hopefully, some day we can meet in person! I would also like to acknowledge Dr. Susan Bondy and Dr. Catherine Gordon for their participation in the defense examination and in the review of my thesis. A huge thank-you goes out to Cayhee Cheung and Shalini Misir who were assigned the difficult task of locating and recruiting the study participants. You both showed incredible patience and perseverance and much of the success of this study is due to your efforts. As well, I would like to thank Cheryl Chase for your commitment to this project and for always being willing to stay late and work weekends to conduct the bone density assessments. Thank you too, to Pilar Furio-Hong for your hard work in setting up and maintaining the database. And then, I want to thank the many people, who, although not directly involved with my thesis, each played an invaluable role in its success. In particular, my husband, Stuart, for your unwavering love and support and your pride in my accomplishments this means more to me than you can ever know. Dr. Susan Jaglal, my MSc supervisor, for continuing to be a great mentor. Sonia Pagura and Helen Lee, with whom I had some of my most enjoyable years in graduate school, teaching in the Department of Physical Therapy; you have both been iv

5 incredibly supportive. I appreciate your friendship, and hope our Thai food nights continue for many years to come! Angela Wall, thank you for always being helpful. I especially enjoyed our many chats! Melissa French, thank you very much for answering my numerous questions every time I popped into your office. You were never too busy to help. Then there was the succession of office-mates over the years, who not only shared my office but also the ups and downs of graduate student life and who were always willing to lend a sympathetic ear: Dr. Joanna Sale, Dr. Jacqueline Hochman and Dr. Cory Borkhoff. Finally, I would like to extend a very special thank you to the study participants. I really appreciated the significant effort you took to be a part of this study. It was truly a pleasure to have met each one of you and I wish you all the very best. Financial Support I gratefully acknowledge the Doctoral Fellowship I received from the Canadian Institutes of Health Research in support of my PhD training. v

6 TABLE OF CONTENTS ABSTRACT... ii ACKNOWLEDGEMENTS... iv CHAPTER 1: Introduction...1 I. Study objectives...1 II. Background...1 A. Osteoporosis Epidemiology and burden of illness...2 Diagnostic classification of osteoporosis...3 Overview of bone metabolism...4 Risk factors for low peak bone mass...5 B. Anorexia nervosa Epidemiology and burden of illness...6 Bone mineral density in women with anorexia nervosa...7 Mechanisms of low BMD in anorexia nervosa...8 Fracture risk in women with anorexia nervosa...10 III. Key determinants of bone mass in women with anorexia nervosa Recovery of bone mass with recovery from anorexia nervosa Effect of illness characteristics on bone mass...17 i. Duration of illness...17 ii. Illness severity...17 iii. Age at onset...18 iv. Subtype Effect of exercise on bone...20 IV. Overview of the present study Study design Life History Calendar interview method Definition of recovery from anorexia nervosa...27 V. Overview of the thesis...28 VI. Significance of the study...30 CHAPTER 2: (Paper 1) The effect of weight and menstrual recovery and other key illness characteristics on bone mineral density in young women with a history of anorexia nervosa...31 Introduction...31 Methods...33 Results...40 Discussion...46 Tables Figures vi

7 CHAPTER 3: (Paper 2) Characterization of changes in bone density associated with illness and recovery in young women with a history of anorexia nervosa...63 Introduction...63 Methods...64 Results...70 Discussion...73 Tables Figures CHAPTER 4: (Paper 3) The effect of exercise during illness and recovery on bone mineral density in young women with a history of anorexia nervosa...86 Introduction...86 Methods...88 Results...94 Discussion...97 Tables Figure CHAPTER 5: Discussion I. Contributions to the literature II. Clinical implications III. Study strengths and limitations IV. Future research V. Conclusions CANDIDATE S ROLE REFERENCES APPENDICES Appendix A: Details of study recruitment Appendix B: Table comparison of characteristics of non-respondents vs. respondents..157 Appendix C: Table characteristics of prior studies investigating recovery of bone mass in women with anorexia nervosa Appendix D: Initial letter of contact Screening form Appendix E: Data collection forms Life History Calendar interview Minnesota Leisure Time Physical Activity Survey OP History and Treatment Questionnaire Calcium Food Frequency Questionnaire Non-respondents telephone interview vii

8 Chapter 1: INTRODUCTION I. Study objectives It is well-established that anorexia nervosa (AN) is associated with decreased bone mineral density (BMD). 1-4 However, current evidence is inconclusive as to whether this reduction in bone mass is permanent or is reversible with sustained recovery from AN This is important to clarify as low BMD associated with AN may not only increase contemporaneous fracture risk in this population, but if these deficits in bone density are sustained even in those who recover, there may also be increased susceptibility to fractures in later life as well. The primary objectives of this study were to determine the extent of reversal of skeletal deficits associated with recovery from AN, and to evaluate the effects of key illness characteristics on BMD, in a cohort of young women who had received inpatient treatment for AN. Secondary objectives were to determine the duration of AN recovery required for normalization of bone mass if complete reversal of skeletal deficits occurred, to examine whether there was a differential capacity for regeneration across skeletal sites, and to elucidate an optimal definition of AN recovery from a bone perspective. II. Background The first part of this chapter will provide an overview of the epidemiology of osteoporosis and of AN, the mechanisms of decreased bone mass in this clinical population and the significance of low BMD both in terms of the impact on the patient and the larger impact on the health care system. 1

9 A. Osteoporosis: epidemiology and burden of illness 2 Osteoporosis (OP) is a systemic skeletal disease characterized by low bone mass and micro-architectural deterioration of bone tissue resulting in reduced bone strength and increased fracture risk. 11 The prevalence of OP in North America among women over the age of 50 is 19 30% It is estimated that 56% of women will sustain a fragility fracture after the age of A fragility fracture is one which is caused by injury that would be insufficient to fracture normal bone. 16 The most common fracture sites are those of the wrist, vertebra and hip which can result in significant diminishment of quality of life, increased mortality rate and considerable health care costs This is particularly true for hip fractures which are associated with more deaths, disability and economic costs than all other OP-related fractures combined. 14, 15, 20 The incidence of hip fracture increases after age 60 and is the most common fragility fracture site in those over 80 years of age. 15 Up to 25% of OP-related hip fracture cases die within the first year, 21, 22 and at one year post-fracture, 40% are unable to walk independently and 60% have difficulty performing at least one essential activity of daily living. 23 Moreover, 17-27% of hip fracture patients are admitted to long-term care facilities for the first time as a direct consequence of the fracture. 18, 19, The estimated annual cost of hip fracture treatment in Canada is $650 million. 22 Together, these statistics highlight the considerable costs of OP both to the health care system and to the patient.

10 Diagnostic classification of OP 3 OP is clinically defined by assessment of bone mineral density (BMD). 11, 24 Dual-energy X-ray absorptiometry (DXA) is accepted as the most accurate clinical method for measuring BMD. 11, 25 BMD is recognized as a continuous risk factor for fracture and thus, no specific fracture threshold exists. 26 However, for diagnostic purposes, the World Health Organization (WHO) proposed two thresholds of BMD based on standard deviations (SD) below the healthy, young adult mean BMD value (ie. after peak bone mass has been achieved). The SDs are reported as T-scores. According to WHO classification criteria, OP is defined as a T-score -2.5, and low bone mass (or osteopenia) as a T-score between -1.0 and -2.5 at the lumbar spine (LSP) or femoral neck (FN). In general, it has been shown that for each SD decrease in BMD, fracture risk approximately doubles. 24 There is also good evidence of high risk of fracture in patients with T-scores -2.5 and a significant reduction in fracture risk with treatment, making this threshold an evidence-based criterion for the diagnosis OP and for treatment initiation. 26 The WHO diagnostic criteria, however, were intended to apply to postmenopausal Caucasian women only, and the applicability of these criteria to other populations is unclear. The use of T-scores, particularly in children who have not yet achieved peak bone mass is inappropriate and meaningless. 26 It is recommended that Z-scores, which are SDs above or below the mean for age, weight, sex and ethnicity, be used for children as well as for women between the ages of years. 26 Z-scores are based on the DXA manufacturer s reference population. At the commencement of the present study there was no recommended criterion for the diagnosis of low bone density in premenopausal women and we used the accepted cut-off value at the time, of a Z-score Recently, the International Society for Clinical

11 Densitometry (ISCD) recommended that a Z-score of -2.0 be considered abnormal. This 4 threshold value is not based on evidence of associated fracture risk, as this has not been established in young women. Rather, it is based on the normal distribution of most biological variables in which a normal range is defined as the normal mean plus or minus 2.0 SDs. 26 Overview of bone metabolism and pathogenesis of bone loss Bone is a dynamic connective tissue which undergoes a process called modeling to shape bones, and a lifelong, continuous process of bone resorption and formation termed remodeling, for bone renewal. 27 Modeling occurs primarily during growth but can also occur in adults to change the shape of the bone in response to mechanical loads. 28 The primary functions of bone remodeling are to maintain mechanical strength by continually replacing old, fatigued bone with new bone, and to maintain mineral homeostasis by providing access to calcium and phosphorus that are stored in the skeleton. 28 Unlike the remodeling process, during bone modeling there is increased bone formation which is not closely linked to prior bone resorption. Conversely, remodeling involves the sequential action of a group of cells, the bone remodeling unit (BRU), including osteoclasts that are responsible for bone resorption and osteoblasts which replace bone that has been removed by the osteoclasts In states of normal bone turnover there is a close coupling of resorption and formation which is essential to prevent bone loss. When this balance is disrupted due to hormonal alterations that occur during the menopause, or as a result of aging, medication use, or illness states such as AN, this can result in serious skeletal diseases, of which OP is the most common. 28 There are two types of bone in the human skeleton cortical bone and trabecular (or cancellous) bone and the rate of remodeling differs by type, reflecting their different functional

12 roles. 28 Approximately 80% of the adult skeleton is comprised of cortical bone, but the 5 proportions of the two types of bone vary among different skeletal sites. The composition of human vertebrae is approximately 80% trabecular bone while the FN is composed of 65% cortical bone. 28, 30 Cortical bone has a much lower turnover rate than trabecular bone, reflecting its primary mechanical role of providing structural support and protection of vital organs. Trabecular bone with its high turnover rate is considered to play an important role in mineral metabolism. 31 Because trabecular bone is more metabolically active than is cortical bone, it may be particularly susceptible to hormonal disturbances Thus, bone loss at the LSP during the menopause is more marked than at the hip, at least in part due to the heightened response of trabecular bone to estrogen suppression. 36 A greater rate of bone loss at the spine may also occur with illnesses such as AN as a result of hormonal disruptions As well, regeneration of trabecular bone may occur at a faster rate than cortical bone with recovery from AN, but this remains uncertain. 6, Risk factors for low peak bone mass Bone loss commences at the time of the menopause and continues until the end of life. 36 Risk of OP and OP-related fractures depends not only on the rate of bone loss but on the amount of bone accrued earlier in life, as BMD at menopause is determined by peak bone mass acquired during adolescence. Factors negatively impacting the accumulation of bone mass during the critical period of rapid skeletal growth, therefore, may have serious consequences in later life. While hereditary factors are important determinants of peak bone mass, accounting for 60-80% of its variance, other factors affecting bone accrual have been identified. In a large Canadian cohort of healthy, premenopausal women, low body weight, older age at menarche,

13 low calcium intake and amenorrhea were negatively associated with BMD at the spine or hip. 51, 52 Other lifestyle factors such as smoking and excessive alcohol consumption have also been 6 evaluated in premenopausal women with conflicting results, but both may be associated with lower BMD. 53, 54 Not only are these risk factors particularly prevalent among women with AN, but onset of AN typically occurs during adolescence, thus putting women with this illness at a significantly higher risk of low peak bone mass. B. Anorexia Nervosa: epidemiology and burden of illness AN is a psychiatric disorder, characterized by a distorted body image and an intense fear of becoming fat, that leads to substantial patient-induced weight loss, progressive malnutrition and primary or secondary amenorrhea. 55 In Ontario, Canada, the prevalence of AN in women aged years has been reported to be 0.56%. 56 This is higher than that reported by a meta-analysis of studies conducted in Europe and the United States, where the average prevalence of AN was found to be 0.3% with an incidence of 8 cases per population per year. 57 Incidence rates are highest for females aged years. In a population-based study in Rochester, Minnesota, USA the incidence rate in this age group was 74 per person years between Data on time trends are conflicting but there is some evidence to suggest that the incidence in the high risk age group (15-24 years) is rising while the incidence in older women has remained stable since AN can cause significant complications in many organ systems including the cardiovascular, gastrointestinal, renal, reproductive, neurologic and skeletal systems. 60 Profound endocrine and metabolic disturbances result from severe malnutrition and purging behaviours leading to amenorrhea, delayed puberty, hypercortisolism and decreased insulin-

14 like growth factor (IGF-1) and leptin levels. 3, 61 The long-term consequences of these medical 7 complications are unknown. 61 Numerous reports have been published on the course and outcome of AN. Although results from these studies are difficult to summarize due to wide variability in the outcome parameters used across studies, there is consistent evidence that AN often follows a variable and protracted course. Several reviews of these studies, in which follow-up periods ranged from 1-29 years, reported that approximately 50% of patients with AN achieve a good outcome (recovery from all essential clinical signs and symptoms of AN), 30% achieve an intermediate outcome (improvement with residual symptoms), and 20% develop a chronic course of illness In general, studies with more stringent recovery criteria have documented lower rates of complete recovery (approximately 25%), finding that there are substantial continuing psychological features of the illness in many patients deemed to have a good outcome. 66 The mortality rate associated with AN is reported to be 0.56% per year or 5.6% per decade (95% CI 3.3%-7.9%), more than 12 times higher than the annual death rate for females years old in the general population and twice that associated with other psychiatric disorders. 67 In studies specifying the cause of death, 54% died from complications of the eating disorder, 27% committed suicide and 19% died of unknown causes. 67 These data emphasize that AN, although relatively rare, is a serious psychiatric disorder with significant morbidity and mortality risks. Bone mineral density in women with AN Since AN typically affects the adolescent female, thus occurring during a critical period of rapid bone accrual, one of the important consequences of AN is a failure to achieve peak bone mass; women with a later onset of AN may experience bone loss. 41 Numerous cross-

15 sectional studies have demonstrated that adolescents and young women with AN have lower BMD at the LSP, 1, 2, 6, 9, 38, 39, 41, 42, hip, 2, 9, 38, 41, 69, 72, 73, 75, 78-81, 83 whole body, 1, 39, 82, 83 and radius6, 9, 71, 86 than healthy age-matched controls. 8 Studies have provided wide-ranging estimates of the prevalence of low bone mass in this population. Differences in results are due primarily to variability in the criteria used to define low bone density: comparison of BMD values to small numbers of age-matched healthy controls; SDs below the young, healthy, adult mean BMD value (T-scores); or, SDs below the mean for weight, age and sex-matched normative data (Z-scores). The Z-score is considered the most appropriate way to report BMD in adolescents and young, premenopausal women. 26 The largest study that used Z-scores included 170 adolescents with AN and reported that 44% and 13.5% had Z-score values < -1 and < -2, respectively, at the LSP, and 25% and 5.9% had Z-score values < -1 and < -2, respectively, at the FN. 38 Smaller studies of adolescents (43 to 85 participants) reported a range in the prevalence of Z-scores of < -2 from 9.1% - 22% at the LSP and 3% - 6.8% at the FN. 3, 39, 72, 75 Only one small study of young adults (44 premenopausal women) reported Z-score values, finding that 50% of participants had Z-scores < -2 at either the LSP or FN. 73 Although the estimates of the prevalence of low BMD varied across the studies, collectively they show reduced bone mass especially at the LSP, indicating preferential loss of trabecular bone which, as described earlier, may be particularly susceptible to hormonal disturbances. Mechanisms of low BMD in AN The underlying mechanisms responsible for low bone mass in AN patients are not fully understood but appear to involve multiple and interrelated endocrine disturbances that occur

16 9 in response to profound undernutrition. 87 In postmenopausal OP the main precipitant of bone loss is increased bone resorption secondary to estrogen deficiency. In estrogen deficient states, osteoclasts are activated by cytokines, such as interleukin-6 and tumor necrosis factor (TNF) which are normally inhibited by estrogen, resulting in increased bone resorption. 88 Estrogen deficiency and amenorrhea are hallmarks of AN: many studies have shown a negative correlation between duration of amenorrhea and BMD. 2, 38, 42, 69, 70, 89 However, estrogen therapy fails to improve bone density in these patients Additionally, patients with AN have lower bone density than age-matched women with hypothalamic amenorrhea from other causes. 93 Together, this suggests that hypoestrogenic status alone is insufficient to explain low BMD in this population. Studies using bone turnover markers have observed an uncoupling of bone turnover with decreased bone formation and increased bone resorption in adult, premenopausal women with AN. 94, 95 In adolescent girls, a decrease in both markers of formation and resorption have been observed, suggesting a state of low bone turnover rather than an uncoupling of the remodeling process. 4, 42 The effect of AN on bone metabolism, therefore, likely differs depending on the age of the patient. 96 A number of nutritionally-regulated hormones have been implicated in the disruption of bone turnover in AN; most of these studies have focused on the effects of insulin-like growth factor (IGF-1), leptin and cortisol. IGF-1, a bone trophic factor and a sensitive index of nutritional status, has consistently been shown to be reduced in AN 42, 75, 78, and to increase rapidly with weight gain accompanied by an increase in bone formation markers. 97, 98, 101 As well, the administration of recombinant human IGF-I to women with AN has been shown to significantly increase bone mass at the LSP. 95 Similarly, low levels of leptin, a hormone involved in the regulation of food intake and energy expenditure, have been demonstrated in

17 AN patients with increased concentrations observed with weight gain, 101, 106 sometimes to excessive levels. 103, 107 Leptin s effects on bone metabolism are not fully understood, but this hormone likely has indirect effects via its influence on the hypothalamic-pituitary-peripheral endocrine axes and may also directly stimulate bone formation Hypercortisolemia has also been reported in women with AN due to increased frequency of secretory bursts and has been found to be associated with decreased markers of bone formation in this population. 70, 106, Investigators have recently identified numerous additional hormonal disturbances in AN with some evidence to suggest that they may be further contributing to the low bone density seen in 10 these individuals. These include increased grehlin 116, osteoprotegerin, 96, 117 and peptide YY, 118 and reductions in dehydroepiandrosterone sulfate (DHEAS), 75, 90, 119 endogenous androgens, 94, 120 and tri-iodothyronine (T3). 94 Fracture risk in women with AN The significance of low BMD is its association with increased fracture risk. In postmenopausal women, fracture risk is approximately doubled for each SD that BMD falls below young adult normal values. 121 Although low BMD has been clearly associated with AN, an association with increased risk for fracture has not been clearly established in the AN population. Rigotti et al 89 reported a relative risk (RR) for non-spine fractures of 7.1 compared with age-matched healthy women. This estimate was based on a sample of only 27 subjects (total follow up 59.8 person years). Lucas et al 122 and Verstergaard et al 123 both used historical data from administrative health registries and compared the occurrence of fractures in patients diagnosed with AN to age-and sex-specific population fracture rates. They reported a RR for all fracture types of 3.0 and 1.98, respectively, in AN patients, and noted that this increase in risk

18 11 persisted for more than 10 years after diagnosis. Vestergard et al 124 pooled their results with those of Rigotti et al 89 and Lucas et al 122 to obtain a pooled RR of 2.6 for all fracture types, 4.7 for symptomatic spinal fractures and 5.3 for hip fractures. There are important issues, however, that were not addressed by the two studies that used health registries: they did not differentiate fractures as low-trauma fragility or traumatic fractures, and did not clarify whether the subjects had chronic AN or had recovered. Furthermore, none of the studies examined asymptomatic spine fractures using x-rays. As symptomatic spine fractures consist of only a small proportion of total spine fractures, the reported RR for spinal fractures may have been underestimated. III. Literature review: key determinants of bone mass in women with AN As previously described, the onset of AN typically occurs during adolescence, the time period of rapid growth when the majority of bone mineral is accrued. AN during this time period may critically impede the attainment of peak bone mass leading to permanent deficits and potentially increasing the risk for OP and fracture in later life, even in those who have recovered from this illness. This underscores the importance of determining the effect of recovery from AN on BMD and identifying key illness characteristics that may influence bone mass in women with a current and past history of AN: identification of risk factors for low BMD in this population will enable the implementation of optimal OP management strategies directed towards those individuals at greatest risk. The next section summarizes the current literature regarding the effect of AN recovery and other key illness characteristics on bone mass in the AN population.

19 1. Recovery of bone mass with recovery from AN 12 There have been a number of studies that have investigated whether BMD deficits are reversed with recovery from AN. (See Appendix C for details of studies). However, their results have been widely conflicting, demonstrating complete recovery of bone mass, partial recovery, no recovery, or recovery at some skeletal sites but not others. One of the earliest studies by Treasure et al 9 reported that BMD values at the LSP, FN and radius (RD) of 25 recovered AN patients were not significantly different from a healthy comparison group. They concluded that bone density returns to normal with recovery from AN. However, a definition of recovery was not provided, duration of recovery was unclear and potential confounders such as duration and severity of the illness were not controlled for in the analysis. Since then, four small studies have reported similar results indicating complete reversal of bone deficits with AN recovery. Bachrach et al 5 demonstrated that BMD values at the LSP and total body (TB) of 9 recovered AN patients were not significantly different from values observed in healthy controls. They defined recovery from AN as having attained a BMI within 1 SD of the mean for healthy, age-matched females. The duration of recovery was not specified. Hay et al 125 found that 21 patients with a good outcome (defined as resumption of menstruation and weight rehabilitation to 85% of average weight for age and sex) had spinal BMD values similar to a control group, whereas patients classified as having an intermediate or poor outcome had significantly lower values. Wentz et al 126 evaluated a sample of 36 patients assessed approximately 11 years after onset of AN and did not find significant differences in the prevalence of low BMD in AN participants compared to healthy controls. However, 42% of the participants had low BMD (defined by the authors as a T-score < -1.0) at the FN compared to 26% of controls. This difference may not have achieved statistical significance due to the small sample size. Moreover, the prevalence of low

20 13 BMD in the control group may be higher than would be expected in young, healthy women. A definition of recovery was not provided. The fourth study, by Bass et al, 127 documented near normal values at the LSP (mean Z-score of -0.4) and normal values at the TB (mean Z-score of 0.2) in 13 patients recovered for 2.7 years. These patients were considered recovered if they had a body weight of 85% of that expected, and regular menses for at least three cycles. Partial recovery of bone mass has been demonstrated in several studies, many of which were prospective studies that may have been of inadequate duration to detect complete normalization of bone density. The largest prospective study by Castro et al 45 followed 108 ill adolescent patients for between 6 and 30 months after admission for inpatient treatment. Sixtyfour patients were classified as having a good outcome at follow-up (BMI > 19kg/m 2 and resumption of menses at the follow-up visit). These patients experienced annual bone mass gains at both the LSP and FN, whereas those classified as having a poor outcome experienced a loss of bone mass at these sites. Despite gains in bone mass, only 17% of patients with a good outcome and low BMD at baseline achieved normal BMD values at follow-up; the authors suggest that in the longer term it may be possible for the majority to reach normal values. Viapiana et al 128 followed 55 patients for 12 months after a treatment program and reported an increase equivalent to 1 SD at the LSP and FN. All patients maintained a BMI of > 17.5 kg/m 2 but it was unclear as to whether they had resumed menstruation. Dominguez et al 129 noted an increase in BMD at the LSP and FN in 28 patients after only 2 months of inpatient treatment, but values remained lower than controls. In a recent 12 month prospective study of 34 AN patients, BMD at the LSP stabilized in 14 patients who experienced a 10% increase in BMI and resumed menstruation for three cycles, compared to those who either gained weight without menstrual recovery or experienced no recovery at all, who continued to experience bone loss. 130 In their

21 14 cross-sectional study, Herzog et al 6 evaluated the long-term outcome of BMD in 51 AN patients approximately 11 years after admission to a treatment facility. They found that 28 patients with a good disease outcome (i.e. regular menstruation and a weight deviation <15% from the expected size-and age-averaged body weight) had significantly higher LSP and RD BMD values than those with a poor outcome, but their values were still lower than those of the healthy reference group. In summary, these studies generally show increases in BMD associated with improvements in weight and/or resumption of menstruation, but deficits were still observed. It is unclear if full recovery of bone mass is possible as no study followed sufficient numbers of recovered patients for a sufficient period of time. In contrast, other studies found no evidence of regeneration of bone mass with AN recovery. Nine prospective studies, which together included only 78 recovered AN patients followed for 3 months to 2 years, found that change in bone density at the LSP and/or TB was not significantly different for recovered versus non-recovered patients. 4, 8, 85, 89, 98, Recovery was variably defined across these studies; three studies defined recovery as weight restoration and resumption of menses, 85, 133, 134 three studies defined recovery by a weight restoration criterion only, 4, 89, 131 two studies by menstrual status only, 98, 132 and one did not specify a definition. 8 Ward et al 10 reported that, in their cross-sectional study of 18 patients with a median duration of recovery of 6 years (recovery defined as a BMI > 18.5 kg/m 2 and resumption of menstruation for at least 6 months), 12 patients had BMD T-scores < An additional crosssectional study compared the lumbar spine BMD of 34 patients with active AN, 20 patients who were weight recovered and amenorrheic, and 19 patients who were both weight recovered and had regular menses for > 3 months (duration of recovery from 3-26 months). There were no significant differences in mean BMD values across the 3 groups; 26% of the weight and

22 15 menstrual recovered patients had BMD values more than 1 SD below the mean of an agematched reference population. 97 Finally, several studies observed differential effects of AN recovery across skeletal sites. Bolton et al 44 observed increases in BMD at the LSP and TB but not at the FN in 15 patients who achieved a normal body weight over the course of 1 year. Miller et al 135 made the observation that patients who resumed menstruation had increased BMD at the LSP but not FN while those who had weight gain only (to > 85 % of ideal weight) had significant increases in BMD at the hip but not LSP, over a mean follow-up period of 13 months. Morris et al 136 examined BMD in a community-based sample of 51 non-recovered AN patients and 36 recovered AN patients (length of recovery ranged from 1 26 years; recovery was not defined) and observed that TB BMD was not significantly different between the recovered group and control group but BMD at the LSP remained significantly lower in the recovered group. Hartman et al 46 reported that in 19 patients recovered beyond clinical dispute for a median of 21 years BMD was similar to controls at the LSP but was significantly lower at the FN. There are many potential explanations for these inconsistent study results. First and foremost are the variable definitions used to define AN recovery across the studies. These included weight restoration only (defined as a > 10% increase in weight, weight to 85% of ideal weight, or a specific BMI cut-point) or weight restoration and resumption of menses, or resumption of menses without consideration of weight restoration. Recently, there has been growing evidence that weight recovery and menstrual recovery may have both combined and independent effects on regeneration of bone mass. 97, 129, 130, 135 If so, the use of variable definitions of AN recovery would contribute to different conclusions regarding the effect of recovery on

23 16 BMD. Many prior studies were also limited by small sample sizes and significant loss to followup in longitudinal studies. This resulted in an inability to rigorously evaluate the effect of AN recovery on BMD while controlling for possible confounding factors such as illness duration and severity, calcium intake, and smoking and alcohol consumption, due to a limited distribution across key variables and inadequate statistical power. In particular, many of these studies did not include enough participants recovered for long periods of time in cross-sectional studies or follow-up participants in prospective studies for long enough to adequately examine the ability of bone mass to fully regenerate with sustained AN recovery. Variability in the durations of recovery or follow-up periods across studies may have also contributed to inconsistent results. Finally, small sample sizes and large drop-out rates also create the potential for selection bias. Overall, these methodological limitations may have contributed to the inconsistent findings reported by these studies and affected the internal validity and generalizability of their results. Consequently, there remain many outstanding issues regarding the effect of recovery from AN on bone mass. It is still unknown whether full regeneration of bone mass occurs at each skeletal site, and, if so, the duration of recovery required. It is also unclear whether the potential for regeneration of bone is impacted by the age at which recovery occurs. Most of the studies were conducted on adolescents and these did not show an obvious difference in results compared to studies that included adult participants. The relative effect of weight restoration versus resumption of menses also requires further evaluation to help establish a definition of AN recovery for future investigations of AN and bone mass.

24 2. Effect of AN illness characteristics on bone mass 17 While recovery from AN may well be the strongest determinant of BMD, certain illness characteristics such as the duration and severity of the illness, age at onset of AN, AN subtype, and exercise may also have important effects on BMD. i. Duration of illness Many studies have demonstrated a negative association between either duration of illness or duration of amenorrhea and BMD in ill AN patients. 1, 2, 38, 39, 41, 42, 69, 77, 79, 90, 97, 125 However, the effect of illness duration on BMD among women who have recovered from AN is less clear. Three cross-sectional studies did not observe a relationship between duration of illness and bone mass in recovered patients, 5, 46, 97 but small sample sizes may have precluded detection of a significant effect. One prospective study reported that illness duration was negatively correlated with change in BMD at the FN but not at the LSP in 19 patients who were followed for 2 years after an inpatient re-feeding program. 8 A definition of recovery was not provided and it is unclear how many of these patients were weight restored and/or resumed menstruation over the 2 years. ii. Illness severity There have been inconsistent results regarding the effect of illness severity on BMD in both ill and recovered patients. This may be due in part to the different variables used to approximate severity of illness. Five studies found no association between BMD and lowest BMI, 10, 81, 85 percentage decrease in body weight, 39 and duration of time at 85% below normal weight. 46 Other studies, however, reported a significant negative correlation between BMD and

25 18 lowest BMI, 74, 79, 125 duration of time at lowest BMI 81, degree of underweight below a BMI of 18.5 kg/m 2, 6 or having ever had a BMI < 15 kg/m An issue that was not adequately addressed in these studies is the potential correlation between illness severity and other illness characteristics; in particular, duration of illness. In the studies that conducted a multivariable analysis including both severity and duration of illness, 6, 38, 81, 125 only duration of illness remained significant, suggesting that illness severity may not be an important independent predictor of low bone mass. Rather, any observable effect of severity may, in fact, be due to a longer illness duration. Further investigation is required to determine the relative effects of chronicity and severity on BMD in both ill and recovered patients. iii. Age at onset of illness Almost half (46.5%) of bone mineral content (BMC) in adulthood is acquired during the 2-year period across peak height velocity, resulting from a combination of bone modeling and remodeling. Two years after menarche, BMC is 85% of the adult value, and 90% of peak bone mass is acquired by age , 138 Thus, it has been hypothesized that onset of AN during adolescence, because of the interruption during this critical period of rapid bone accrual, would result in a greater deficit in bone density than would adult onset AN. 41, 127 While this makes biological sense, most studies have not demonstrated an association between age at onset of AN and BMD, although their findings were limited by small sample sizes and narrow ranges in onset ages. 6, 10, 46, 68, 69, 71, 77, 82, 125, 139 Given the underlying hypothesis, it may be preferable to evaluate the effect of age at onset on bone mass in terms its relation to onset of puberty, and this has rarely been addressed. Two studies that compared bone density in patients with primary versus secondary amenorrhea did not find significant differences between the two groups, but

26 19 were limited by small numbers of patients in one or both groups. 39, 126 A third study reported that patients with primary amenorrhea had lower BMD at the spine and hip, but these differences were not significant after adjusting for duration of illness. 41 More research, which specifically evaluates the effect of premenarcheal versus postmenarcheal onset of AN on BMD, is needed. iv. Subtype of AN The DSM-IV sub-divides AN into two types restricting type (R) and bingeeating/purging type (BP). The R subtype describes presentations in which weight loss is achieved primarily through dieting or fasting whereas the BP subtype achieves weight loss by regularly engaging in purging behaviours, such as self-induced vomiting, misuse of laxatives, diuretics or enemas. 55 It has been suggested that the BP type may be associated with lower BMD due to greater disturbances in the metabolic and endocrine systems. 85 Yet, there is minimal evidence to support that AN subtype is an important determinant of bone density. One prospective study reported that, after adjusting for body weight, age at onset and duration of illness, subtype of AN was the best predictor of spinal BMD at follow-up, with the BP subtype associated with significantly lower BMD. 85 However, only 7 and 8 patients of the R and BP subtypes, respectively, completed the study, raising the possibility that this may have been a spurious finding. Two cross-sectional studies also reported that patients with BP subtype had lower BMD, but the difference between the two groups was not statistically significant. 68, 74 Small subtype groups in these two studies may have precluded detection of significant differences. Conversely, several studies found that BP subtype was associated with higher BMD, although this may have been due to higher body weight in the BP participants. 71, 77, 78, 140

27 20 Only one study appeared to adjust for weight and noted that, once weight was controlled for, the effect of subtype was no longer significant. 78 Finally, one study reported that frequency of vomiting predicted spinal BMD, but it is unclear whether it was associated with higher or lower bone mass. 69 Taken together, AN subtype seems unlikely to be a key predictor of bone mass and any effect may be due to differences in body weight or other illness characteristics. A study which controls for these factors is needed to confirm this. In summary, the majority of these studies have been limited by small sample sizes and by inadequate consideration of important confounding factors, leading to inconsistent results. Many questions, therefore, remain regarding the effect of these illness characteristics on bone mass in women with a history of AN. Not only have the impact of these characteristics on BMD in ill patients not been clearly elucidated, but there is sparse evidence about the extent to which possible negative effects persist in women who have recovered from this eating disorder, or the extent to which prior illness history influences the rate of bone regeneration which may occur with AN recovery. 3. Effect of exercise on bone Although exercise could be considered another key illness characteristic in women with AN, it is a factor that deserves particular consideration. Excessive or compulsive exercise is a salient feature of this disease and is believed to play a role not only in the progression and maintenance of AN but in its pathogenesis as well Consequently, standard practice in the treatment of eating disorders is to proscribe exercise. 145, 146 Yet, very little is known about the effect of exercise on bone in ill patients and even less about the effect of exercise during recovery from AN.

28 21 Physical activity imparts mechanical loads on bone, initiating an adaptive structural response which involves increases in bone mass and changes in geometry that may ultimately enhance bone strength. 147 Exercise in healthy women has consistently been shown to have a positive effect on bone density. 148 Activities that impart both a high strain rate (rapid deformation of bone in response to mechanical load) resulting from dynamic and atypical movement directions, and a high force magnitude, have demonstrated the greatest osteogenic stimulus Specifically, sports such as gymnastics, dance, volleyball, tennis and soccer have been found to have the most substantial effect on bone mass while minimal bone loading activities such as walking, swimming or cycling have no effect. 149, 150, The response of bone to physical activity in women with AN, however, has not been so clearly established. States of severe nutritional deficiency and subsequent disturbances of the endocrine environment may alter the anabolic response to mechanical loading on the skeleton such that greater loads may be required to trigger an adaptive response. 158 Alternatively, excessive exercise during illness may attenuate any positive effects by contributing to further weight loss. In contrast to periods of malnutrition, during recovery an adaptive response may be stimulated by lower mechanical loads due to increases in estrogen and IGF and the possible rapid regeneration of bone mass that occurs during recovery from AN. Prior studies of BMD and physical activity in women with AN have demonstrated marked inconsistency of results. Ten studies found no association between exercise and BMD. 1, 6, 69, 70, 76, 80, 83, 89, 133, 159 These were small studies ranging from participants (average of 37 participants). In contrast, three additional studies demonstrated moderate evidence of a positive effect of exercise on BMD. 38, 41, 75 In a study of 170 adolescent AN patients, greater than 3 hours per week of exercise was associated with higher spinal BMD, but this association was not

29 22 significant after adjusting for duration of illness and body weight. 38 Gordon et al 75 observed, in their sample of 61 adolescent patients, that hours per week of current exercise participation was positively correlated with BMD at the LSP, FN and TB. Seeman et al, 41 however, discovered significantly higher BMD at the FN, but not at the LSP, in participants who participated in sport at least 2-3 hours per week for 2-3 hours each time. Finally, two studies indicated that exercise may have a negative effect on bone mass. 44, 139 Bolton et al 44 stated that high exercise was correlated with lower spinal BMD. The definition of high exercise was unclear that is, whether it was based on the amount of exercise or the type of bone-loading. The second study reported that 1-6 hours/week of exercise was positively associated with BMD at the hip and spine but that performing no exercise or exercising > 6 hours/week was negatively associated with both skeletal sites. 139 However, these findings must be questioned as they were based on only 8 AN participants. These conflicting results are undoubtedly due, at least in part, to the variability in the way exercise was assessed. Most of these studies had limited exercise data and evaluated the effect of current activity only. Many studies measured only intensity of exercise based on hours per week or estimates of caloric expenditure and did not take into consideration the osteogenic potential of specific activities. The type of exercise may be more relevant to the evaluation of bone response than is the amount and intensity of exercise. Furthermore, rather than considering only recent exercise, it may be important to determine the impact of exercise history throughout life, especially exercise performed during adolescence, which may confer a lifetime protective effect. 160, 161